雲端運算合作夥伴計劃
Mirendil與Google Cloud簽訂逾1億美元運算容量協議
Mirendil 宣布與 Google Cloud 建立多年期合作夥伴關係，為其自我改進型 AI 研究鎖定超過 1 億美元的算力資源。這家由前 Anthropic 研究人員創立的 AI 實驗室，將可使用 Google TPU、Nvidia GPU 及託管訓練集群，開發能夠迭代改善自身知識與表現的系統。
是次合作安排，讓 Mirendil 可靈活地將不同工作負載配對至不同加速器，同時運用自家軟件及系統層，提升硬件使用效率。Google 將此合作定位為跨異構基礎設施的大規模協同調度，而 Mirendil 則計劃將算力應用於橫跨 AI、醫學、生物學及材料科學等領域的研究。
對研究人員及企業用戶而言，是次合作有望透過工作負載與硬件的優化配對，實現更快速的實驗及更高效的訓練。這宗協議亦反映出雲端服務商之間，為爭取前沿 AI 實驗室以大型基礎設施承諾作籌碼的競爭日趨激烈。
英文原文
Mirendil announced a multiyear partnership with Google Cloud to secure more than $100 million in compute capacity for its self-improving AI research. Founded by former Anthropic researchers, the AI lab will gain access to Google TPUs, Nvidia GPUs and managed training clusters as it develops systems capable of iteratively improving their own knowledge and performance.
The arrangement gives Mirendil flexibility to match different workloads with different accelerators while using its own software and systems layer to improve hardware utilization. Google positioned the collaboration around large-scale orchestration across heterogeneous infrastructure, while Mirendil plans to apply the compute to research spanning AI, medicine, biology and materials science.
For researchers and enterprise users, the partnership could enable faster experimentation and more efficient training through workload-to-hardware optimization. The deal also reflects growing competition among cloud providers to secure frontier AI labs with large infrastructure commitments.
- 來源
- Trend Hunter
- 發布
- 2026-08-11
- 品類
- Newventures
- 出處
- techcrunch, mirendil